الفهرس Only 14 pages are availabe for public view
 |  | from | 128 |  |  |
| | | from | 128 | | |
Abstract
Surface texture is an important factor that affects both the performance and appearance of machined surfaces. Classification of texture pattern and estimation of roughness parameters have proved to be very important in the evaluation of expected part performance.
In this thesis we present a system using an entry computer data logging visible unit (ECDVU) with an artificial neural network (ANN) for identification and classification of engineering surfaces and roughness estimation. The methodology implemented in a software rendering based on stored surface roughness data measured by Form Talysurfseries2as a stylus instrument and captured images. So the objective of this work is the utilization of six ANNs to classify some kinds of machined surfaces and estimation its surface roughness parameters without contact or any knowledge about machining conditions (cutting speed, feed rate, and depth of cut).
With the idea of image processing based on captured intensity images of these specimens six features extraction techniques are applied to obtain inferred parameters for all machined surfaces.
The six classifiers are presented related to the used extracted features techniques.
The feature extraction techniques considered in the study are namely: grouped features, gray level histogram, edge detection, co-occurrence matrices, mean-variance of intensity images, and wavelet transformatiol).
The feed forward back-propagation using adaptive learning rate with momentum term algorithm (BP A) is chosen to activate the ANN to be used as a supervised classifier.
These features are described and compared at the training mode and recalling mode of the ANN to evaluate the ability of ANNs to classify the machined surfaces and estimate the surface roughness parameters and the parameters of the machining process used to produce the workpiece. The output of the ANN classifier is also used to establish the relationship between measured surface roughness parameters and the features extracted from the surface images.
The methodology receives 42 specimens produced by some of the most used machining processes. The five machining processes (groups) namely are, turning, grinding, milling, lapping, and shaping as input, then a precise stylus instrument is used to measure texture features for each one. Four roughness parameters from these measured values are extracted as the average roughness (Ra), root mean square roughness, ( Rq), skewness (Rsk), and kurtosis (Rku).
Sets of calculated amplitude parameters are calculated for the co-occurrence matrices of each gray scale digital image. These calculated parameters are the angular second moment, the contrast, and the correlation. A group includes the measured and calculated feature parameters are used as an input to the first classifier. The other extracted features are used as inputs to the other ANNs classifiers
The outputs of the classifiers are used in estimation of the roughness parameters and one machining parameter, the feed rate. The suggested system gives accuracy between 91.66 % and 100 % in classification of the machined work-pieces. It gives a maximum error in Ra roughness estimation less than value 8.596E-006 Ilm for the images when using lOX microscope objective in image capturing, and a maximum error in Ra roughness estimation for less than value 7.591E-006 Ilm for the images which has Ra
<2.00E-6 Jlm, and less than value 8.5047E-004 Jlm for the images which has R,,>8.00E-6 Jlm when using 5X microscope objective. The maximum error in feed rate estimation is less than value 0.00739 mm/stroke.
So the ANN can be used as a supervised classifier of machined surfaces and accurate estimators of feed rate and Ra roughness value. The results at least make the classification and estimation more trustworthy.